Fighting Disintermediation: Can AI Finally Turn the Tide?
Fighting Circumvention with AI Today, and Preparing for Agentic Commerce Tomorrow
Before we get started, a quick note. I owe a small apology for the long silence between posts. When I launched this newsletter, I’d hoped to publish something every couple of weeks, but life had other plans. Between family, advising early-stage startups, and thinking through my own next chapter, I’ve been busier than expected.
That said, I’ve continued to collect ideas and draft stories I’m excited to share. This post is one I originally queued up a few months ago. But in that short time, AI’s trajectory and the conversation around it has shifted dramatically. Coming back to it, I realized the original framing no longer fit today’s agentic landscape. So I’ve revisited the piece, expanding it with new thinking about AI’s role in either reinforcing or potentially replacing marketplaces as we know them. I focus in particular on a long-standing challenge for these platforms: circumvention.
Thanks for reading and I look forward to your notes and comments.
Picture a guest and a host on Airbnb. They’re on the brink of finalizing a booking, then one party shares a phone number that slips through the content scrubber. Communications on the platform go silent, and poof, the deal is completed offline. This single act lowers the platform’s revenue, damages trust, and makes future transactions harder to control. Many marketplace founders I advise share frustrations about this pattern, often termed “disintermediation.” It’s a problem as old as online marketplaces themselves.
The key questions to understand are:
how much value does your platform need to create to justify keeping transactions within the marketplace?
In which cases is each party incentivized to transact off-platform?
Is disintermediation always nefarious, or are there innocent reasons for it?
One thing is certain: disintermediation will never be zero, no matter what you do. It’s a constant game of whack-a-mole that requires a mix of “carrots and sticks” to reinforce the desired behavior. Marketplaces that succeed in minimizing it often combine pricing strategies, monetization experiments (such as subscriptions), a forced transition to a buyer-commit model, and value-added benefits like payment guarantees or insurance coverage. But these traditional tactics only go so far.
That’s where AI presents an entirely new frontier.
Why Disintermediation Happens
Most marketplaces rely on commissions or take rates for revenue. When users spot a chance to dodge fees, they may do so, especially for high-dollar, long-duration, or repetitive exchanges. Airbnb and Upwork serve as prime examples:
Airbnb charges a 14–16% guest fee and a 3% host fee. A study published by UC Irvine and University of Virginia business graduate programs revealed that 18% of guest-host interactions led to offline bookings. High-value rentals are most at risk.
Upwork scans chat messages for contact information or payment terms, warning or suspending violators. Even then, a conservative estimate of 7% of freelancers that bypass the customer may have cost the company as much as $84M in lost commissions based on 2023 FY extrapolations.
Both cases highlight a core issue: if a platform doesn’t provide enough value, users will find ways around it. Airbnb Plus, the on-platform certification program the company introduced to verify its highest quality listings, reduced offline bookings by about 6% more than strict contact-blocking alone. This highlights the impact of positively reinforcing trust signals over punitive restrictions. That’s not a simple solution to ship, of course.
So why aren’t there more tried-and-true product features, or even SaaS solutions, that meaningfully curb this behavior? The simple answer: marketplaces are HARD. Each one is different, and what works for Platform A might not apply to Platform B. Marketplace teams have gone on countless crusades to tackle disintermediation, but the challenge is compounded by opportunity cost. Given the choice, most leadership teams would rather focus on growth experiments with clear ROI than chase the nebulous impact of reducing funnel leakage from circumvention. As a result, disintermediation roadmaps often take a backseat, much to the frustration of Trust & Safety teams (I may be speaking from experience on this one).
Should we ever expect to see truly game-changing solutions? Maybe this isn’t a battle every marketplace needs to fight alone. AI advancements are opening the door to scalable, automated solutions.
AI Tactics for Reducing Disintermediation
1. Smarter Detection with Large Language Models (LLM)
For years, marketplaces relied on rule-based filters to catch messages containing personal details or hints at offline transactions. Airbnb famously developed regex-based scrubbing tools to block phone numbers and email addresses from being shared before a transaction was finalized. The problem? Users always found creative ways to work around these filters.
Most people already know this part: Enter LLMs. Instead of rigid keyword-matching, LLMs use Natural Language Processing (NLP) to interpret context, making them much harder to trick. They can scan peer-to-peer messaging, detect circumvention attempts, and flag high-risk transactions with far greater accuracy. By training AI models on user data and unstructured data sources, marketplaces can pinpoint areas where disintermediation is most likely to occur without frustrating legitimate users. That’s a major first step in this process.
2. Automated Decision-Making and Enforcement
Once you flag risk, what happens next? Historically, high-growth marketplaces had two options:
Hire more Trust & Safety agents to review flagged messages as content outstrips the team’s capacity.
Engage a BPO (Business Process Outsourcing), a costly and training-intensive choice feasible only for the largest platforms.
Processes with decision trees are made and evolved to take action on circumvention. But without a smart risk-ranked algorithm and consistently applied moderation of user content, these manual reviews are time consuming and largely ineffective for marketplace Ops teams.
Now, Agentic AI changes the game. AI agents, with human-in-the-loop oversight, can analyze each case and apply an appropriate response. This can range from a gentle warning to an account suspension. Over time, this feedback loop refines the system’s risk model, reducing false positives and improving enforcement accuracy. Trust & Safety teams can now scale their coverage without ballooning headcount costs. Some startups are building these tools natively with the help of low-code / no-code admin app builders like Superblocks.
Today’s agentic AI systems offer dramatic gains in scaling trust & safety workflows. But these capabilities still require deliberate human-in-the-loop oversight, especially for edge or high-stakes cases which come with the territory if your marketplace connects two unknown parties to transact. Real-time review remains essential to limit unintended actions and maintain governance standards.
Sidebar: In practice, marketplace teams should treat agentic tools as augmenters, not replacements: human reviewers focus on exceptions and strategic decisions, while AI handles consistent, high-volume enforcement. This hybrid model helps maintain user trust without reverting to full manual triage. That is the audacious scaler at this time and its ok that its not fully autonomous.
One company leading the way here is SafetyKit. Instead of legacy keywords or partial ML screening, SafetyKit’s LLM-based agents process 100% of Upwork job posts, matching against platform-specific policies and workflows. The result? A 95% accuracy rate that’s well above the ~80% human benchmark. This leads to $1M/year in savings and faster policy iteration cycles. This kind of AI-native T&S tooling removes a messy, manual bottleneck and creates a runway for real-time disintermediation enforcement to become scalable and continuous.
3. AI Intermediaries for Negotiations
Many transactions require back-and-forth communication before both parties commit. This is the norm in most double-commit marketplaces (where both supply and demand must agree before a transaction happens).
Here’s where AI-powered negotiators step in.
Instead of exposing personal details, an AI agent can facilitate pricing discussions, add-on selections, and deal structuring within the marketplace itself. Take Peerspace as an example:
A host might set a minimum and maximum pricing threshold, as well as permissible add-ons.
A guest might have a strict budget but need specific services (e.g., catering included for under $1,500).
An AI intermediary can automatically align incentives, suggest trade-offs (e.g., reducing the guest list in exchange for a lower price), and present an optimized offer to both parties.
In the ideal scenario, the AI agent has done the heavy lifting and reached consensus. The transaction is secured out of convenience (big-time value driver).
Considering how to implement this “negotiator” agent? I found this interesting: new research shows that how AI negotiators “sound” matters. In a March 2025 study where LLM agents engaged in 120,000 negotiation rounds, “warmth” correlated with more frequent deals and enhanced ongoing relationships—but at the cost of lower individual-value claims. Conversely, more dominant agents extracted higher immediate value. This opens a design choice for a marketplace: should a platform’s agents prioritize volume and retention (warm tone), or revenue per deal (assertive tone)? Implementing dual-tier agent personas (e.g. one with conciliatory language for first-time or trust-building offers, another more assertive for upsells) may help balance revenue and relational goals. But this still feels like a rabbit hole only the most forward thinking startups and “superplatforms” are ready to go down at this point.
Zooming back out: Now, imagine each side having their own AI agent, negotiating in real-time to reach a fair deal. This approach removes friction, eliminates unnecessary contact information exchanges, and keeps transactions within the marketplace. Feels pretty utopian, doesn’t it?
The Future of AI-Driven Marketplace Integrity
We are at the forefront of AI-powered solutions like these, yet Agentic AI workflows are not yet commonplace. That will change in the back half of this year and early 2026. Early-stage marketplace founders I advise are already testing AI-driven approaches to disintermediation, and as adoption grows, the benefits will be too large to ignore:
LLM-driven scanning offers smarter detection.
Agentic AI enables automated enforcement, making marketplace Ops teams far more productive.
AI negotiators streamline deals, keeping discussions (and transactions) on-platform.
These capabilities will help today’s marketplaces win, but they’re just the beginning. The real disruption may come when AI agents evolve from enforcement tools to autonomous actors in the market. Let’s look ahead at how that might reshape the disintermediation problem entirely.
Envisioning a Parallel Future
Picture this scenario: AI agents not just fighting off circumvention, but acting as competitive gatekeepers, steering users away from marketplaces entirely. Perhaps first-party LLM tools (ChatGPT, Claude, Perplexity) truly become the meta-aggregators for everything?
In the emerging “agentic economy,” assistant agents (on behalf of users) and service agents (on behalf of businesses) may transact directly, bypassing centralized platforms. Disintermediation becomes a potential extinction event for today’s marketplaces. Seem viable?
Hard to say for certain, but marketplaces still hold structural advantages:
Liquidity & Network Effects: AI agents need scale—both supply and demand—to match offers and provide the variety that drives efficient growth across markets.
Search cost reduction: Marketplaces aggregate signal, trust, and discovery better than ad-hoc one-off agent-to-agent matches.
Trust infrastructure: Platforms offer identity verification, dispute resolution, insurance, refund guarantees, and customer support. These are non-negotiables for users transacting real goods & services who are also wary of anonymized bots.
Marketing & reach: Marketplaces amplify visibility and ratings in ways fragmented agent networks cannot easily replicate.
Still, we should sequence this strategically: assuming everyone will inevitably leverage an agent to take undesirable tasks off their plate, marketplace operators should begin prototyping agent-friendly integrations. Exposing structured APIs (e.g. agent-to-marketplace negotiation protocols), enabling agent discovery widgets, or offering agent-certified deals ensures that AI agents enhance rather than fragment the ecosystem. It goes beyond defending turf. Make the platform undesirable for agents to bypass without losing those foundational benefits. The “carrot” approach might just work for robots too!
As we shift toward an agentic economy, where AI assistants negotiate and transact autonomously, marketplaces face a strategic inflection point. Build agent-aware systems from the ground up or risk becoming obsolete. Platforms that sequence LLM-driven negotiation, enforcement, and certification systems alongside agent-friendly APIs will not just defend revenue, they’ll become integral infrastructure in the next wave of digital commerce.
Turning the Tide And Staying Ahead
Disintermediation will always be a challenge, but with the right AI strategies, I believe it can be reduced to a manageable level. First, marketplaces should rejoice in the early wins that today’s AI tools can deliver. Second, they should start planning for what happens as users begin working with surrogate agents. The marketplace winners of the next decade will be those that embrace these innovations, not just to protect their revenue, but to create seamless, frictionless, repeatable, and reliable user experiences. AI won’t eliminate disintermediation, but it may redefine who owns the transaction entirely. The marketplaces that stay ahead will be those that evolve from gatekeepers into orchestration layers that serve both humans and agents, without losing what made the platform valuable in the first place.
It’s time to stop playing whack-a-mole and start leveraging AI to build marketplaces that users (and soon, their autonomous stand-ins) want to stay in. Now the question is which marketplaces are going to lead the way there. Maybe fighting disintermediation and product led growth initiatives can coexist on the roadmap?
Let’s continue the conversation in the comments below.